CSE 573 P: Artificial Intelligence
52
CSE 573 P: Artificial Intelligence Hanna Hajishirzi slides adapted from Dan Klein, Pieter Abbeel ai.berkeley.edu And Dan Weld, Luke Zettlemoyer
Transcript of CSE 573 P: Artificial Intelligence
CSE 573 P: Artificial Intelligence
Hanna Hajishirzi
slides adapted from
Dan Klein, Pieter Abbeel ai.berkeley.edu
And Dan Weld, Luke Zettlemoyer
Search
o Agents that Plan Ahead
o Search Problems
o Uninformed Search Methods
oDepth-First Search
o Breadth-First Search
oUniform-Cost Search
Agents that Plan
Reflex Agents
o Reflex agents:o Choose action based on current percept
(and maybe memory)
o May have memory or a model of the world’s current state
o Do not consider the future consequences of their actions
o Consider how the world IS
o Can a reflex agent be rational?
Video of Demo Reflex Optimal
Video of Demo Reflex Odd
Planning Agents
o Planning agents:o Ask “what if”
o Decisions based on (hypothesized) consequences of actions
o Must have a model of how the world evolves in response to actions
o Must formulate a goal (test)
o Consider how the world WOULD BE
o Optimal vs. complete planning
o Planning vs. replanning
Video of Demo Replanning
Video of Demo Mastermind
Search Problems
Search Problems
o A search problem consists of:
o A state space
o A successor function
(with actions, costs)
o A start state and a goal test
o A solution is a sequence of actions (a plan) which transforms the start state to a goal state
“N”, 1.0
“E”, 1.0
Search: it is not just for agents
Hardware
verification
Search: It’s not just for Agents
11
Hardware verificationPlanning optimal repair
sequences
Search: It’s not just for Agents
11
Hardware verificationPlanning optimal repair
sequences
Planning optimal
repair sequences
Route
Planning
o Search: Modeling the world
E xample: Traveling in Romania
o State space:
o Cities
o Successor function:
o Roads: Go to adjacent city with
cost = distance
o Start state:
o Arad
o Goal test:
o Is state == Bucharest?
o Solution?
What’s in a S tate Space?
o Problem: Pathing
o S tates: (x,y) location
o Actions: NSE W
o Successor: update location
only
o Goal test: is (x,y)=E ND
o Problem: Eat-All-Dots
o S tates: {(x,y), dot
booleans}
o Actions: NSE W
o Successor: update
location and possibly a dot
boolean
o Goal test: dots all false
The world state includes every last detail of the environment
A search state keeps only the details needed for planning (abstraction)
S tate Space S izes?
o World state:
o Agent positions: 120
o Food count: 30
o Ghost positions: 12
o Agent facing: NSE W
o How many
o World states?
120x(230)x(122)x4
o S tates for pathing?
120
o S tates for eat-all-dots?
120x(230)
S tate Representation
o Real-world applications:
o Requires approximations and heuristics
oNeed to design state representation so that search is feasible
oOnly focus on important aspects of the state
oE .g., Use features to represent world states
Safe Passage
o Problem: eat all dots while keeping the ghosts perma-scared
o What does the state space have to specify?
o (agent position, dot booleans, power pellet booleans, remaining scared time)
S tate Space Graphs and Search Trees
S tate Space Graphs
o State space graph: A mathematical
representation of a search problem
o Nodes are (abstracted) world configurations
o Arcs represent successors (action results)
o The goal test is a set of goal nodes (maybe only
one)
o In a state space graph, each state occurs
only once!
o We can rarely build this full graph in
memory (it’s too big), but it’s a useful idea
Search Trees
o A search tree:
o The start state is the root node
o Children correspond to successors
o Nodes show states, but correspond to PLANS that achieve those states
o For most problems, we can never actually build the whole tree
“E”, 1.0“N”, 1.0
This is now / start
Possible futures
S tate Space Graphs vs. Search Trees
S
a
b
d p
a
c
e
p
h
f
r
q
q c G
a
qe
p
h
f
r
q
q c G
a
S
G
d
b
pq
c
e
h
a
f
r
We construct both on demand – and we construct as little as possible.
Each NODE in in the search tree is an entire PATH in the state space
graph.
Search TreeState Space Graph
State Space Graphs vs. Search Trees
S G
b
a
Consider this 4-state graph: How big is its search tree (from S)?
State Space Graphs vs. Search Trees
S G
b
a
Consider this 4-state graph:
Important: Lots of repeated structure in the search tree!
How big is its search tree (from S)?
s
b
b G a
a
G
a G b G
… …
Tree Search
Search Example: Romania
Searching with a Search Tree
o Search:
o Expand out potential plans (tree nodes)
oMaintain a fringe of partial plans under consideration
o Try to expand as few tree nodes as possible
General Tree Search
o Important ideas:o Fringe
o Expansion
o Exploration strategy
o Main question: which fringe nodes to explore?
Recap: Search
o Search problem:o States (configurations of the world)
o Actions and costs
o Successor function (world dynamics)
o Start state and goal test
o Search tree:o Nodes: represent plans for reaching states
o Search algorithm:o Systematically builds a search tree
o Chooses an ordering of the fringe (unexplored nodes)
Search Algorithms
o Uninformed Search Methods
oDepth-First Search
o Breadth-First Search
o Uniform-Cost Search
o Heuristic Search Methods
o Best First / Greedy Search
o A*
Depth-First Search
Depth-First Search
S
a
b
d p
a
c
e
p
h
f
r
q
q c G
a
qe
p
h
f
r
q
q c G
a
S
G
d
b
pq
c
e
h
a
f
rq
p
h
fd
b
a
c
e
r
Strategy: expand a deepest node first
Implementation: Fringe is a LIFO stack
Search Algorithm Properties
Search Algorithm Properties
o Complete: Guaranteed to find a solution if one exists?
o Optimal: Guaranteed to find the least cost path?
o Time complexity?
o Space complexity?
o Cartoon of search tree:
o b is the branching factor
o m is the maximum depth
o solutions at various depths
o Number of nodes in entire tree?
o 1 + b + b2 + …. bm = O(bm)
…b
1 node
b nodes
b2 nodes
bm nodes
m tiers
Depth-First Search (DFS) Properties
o What nodes DFS expand?
o Some left prefix of the tree.
o Could process the whole tree!
o If m is finite, takes time O(bm)
o How much space does the fringe take?
o Only has siblings on path to root, so O(bm)
o Is it complete?
o m could be infinite, so only if we prevent
cycles (more later)
o Is it optimal?
o No, it finds the “leftmost” solution, regardless
of depth or cost
…b
1 node
b nodes
b2 nodes
bm nodes
m tiers
START
GOALa
b
Breadth-First Search
Breadth-First Search
S
a
b
d p
a
c
e
p
h
f
r
q
q c G
a
qe
p
h
f
r
q
q c G
a
S
G
d
b
p q
c
e
h
a
f
r
Search
Tiers
Strategy: expand a shallowest node first
Implementation: Fringe is a FIFO queue
Breadth-First Search (BFS) Properties
o What nodes does BFS expand?
o Processes all nodes above shallowest
solution
o Let depth of shallowest solution be s
o Search takes time O(bs)
o How much space does the fringe
take?
o Has roughly the last tier, so O(bs)
o Is it complete?
o s must be finite if a solution exists, so yes!
o Is it optimal?
o Only if costs are all 1 (more on costs later)
…b
1 node
b nodes
b2 nodes
bm nodes
s tiers
bs nodes
BFS
Algorithm Complete Optimal Time Space
DFS w/ Path
Checking
BFS
Y N O(bm) O(bm)
Y Y* O(bs) O(bs)
…b
1 node
b nodes
b2 nodes
bm nodes
d tiers
bs nodes
Quiz: DFS vs BFS
o When will BFS outperform DFS?
o When will DFS outperform BFS?
Iterative Deepening
o Idea: get DFS’s space advantage with
BFS’s time / shallow-solution
advantages
o Run a DFS with depth limit 1. If no
solution…
o Run a DFS with depth limit 2. If no
solution…
o Run a DFS with depth limit 3. …..
o Isn’t that wastefully redundant?
o Generally most work happens in the lowest
level searched, so not so bad!
…b
Cost-Sensitive Search
BFS finds the shortest path in terms of number of actions.It does not find the least-cost path. We will now covera similar algorithm which does find the least-cost path.
START
GOAL
d
b
pq
c
e
h
a
f
r
2
9 2
81
8
2
3
2
4
4
15
1
32
2
How?
Uniform Cost Search
Uniform Cost Search
S
a
b
d p
a
c
e
p
h
f
r
q
q c G
a
qe
p
h
f
r
q
q c G
a
Strategy: expand a
cheapest node first:
Fringe is a priority queue
(priority: cumulative cost)S
G
d
b
pq
c
e
h
a
f
r
3 9 1
16411
5
713
8
1011
17 11
0
6
39
1
1
2
8
82
15
1
2
Cost
contours
2
…
Uniform Cost Search (UCS) Properties
o What nodes does UCS expand?
o Processes all nodes with cost less than cheapest
solution!
o If that solution costs C* and arcs cost at least , then the
“effective depth” is roughly C*/
o Takes time O(bC*/) (exponential in effective depth)
o How much space does the fringe take?
o Has roughly the last tier, so O(bC*/)
o Is it complete?
o Assuming best solution has a finite cost and minimum
arc cost is positive, yes!
o Is it optimal?
o Yes!
b
C*/ “tiers”c 3
c 2
c 1
Uniform Cost Issues
o Remember: UCS explores increasing cost contours
o The good: UCS is complete and optimal!
o The bad:o Explores options in every “direction”
o No information about goal location
o We’ll fix that soon!
Start Goal
…
c 3
c 2
c 1
Video of Demo Empty UCS
Video of Demo Maze with Deep/Shallow Water --- DFS, BFS, or UCS? (part
1)
Video of Demo Maze with Deep/Shallow Water --- DFS, BFS, or UCS? (part
2)
Video of Demo Maze with Deep/Shallow Water --- DFS, BFS, or UCS? (part
3)
The One Queue
o All these search algorithms are
the same except for fringe
strategies
o Conceptually, all fringes are priority
queues (i.e. collections of nodes
with attached priorities)
o Practically, for DFS and BFS, you
can avoid the log(n) overhead from
an actual priority queue, by using
stacks and queues
o Can even code one implementation
that takes a variable queuing object
Search and Models
o Search operates over
models of the world
o The agent doesn’t
actually try all the plans
out in the real world!
o Planning is all “in
simulation”
o Your search is only as
good as your models…
To Do:
o Try python practice (PS0)
oWon’t be graded
o PS1 on the website
o Start ASAP
o Submission: Canvas
o Website:
oDo readings for search algorithms
o Try this search visualization tool
ohttp://qiao.github.io/PathFinding.js/visual/
